Released HFO scores reduced mean log loss for later detector agreement in 18 SEEG records with a fixed time split
We tested whether released HFO scores improve prediction of agreement between two detectors in later SEEG recording time. We used public OpenNeuro ds005398 version 1.1.3 and all 18 SEEG participants. We fitted separate patient models on the first 29.5 minutes and tested them on minutes 30 to 60. The baseline used event duration and earlier channel agreement. The expanded model added the released artifact and spike scores. Among 117,662 test MNI marks, the mean patient log loss fell from 0.09270 to 0.08519 nats per mark. The mean gain was 0.00751 nats, with a patient bootstrap 95 percent interval of 0.00392 to 0.01202. Gain was positive in 17 of 18 patients. All planned checks had positive mean gain. Separate label and optimizer checks passed, and an offline repeat matched the results. The package preserves code, pinned source bytes, predictions, fitted models, provenance, and repeat instructions. Detector agreement does not show event truth or clinical benefit. The source score training overlap is unknown.
high-frequency oscillations · SEEG · detector agreement · logistic regression · log loss · OpenNeuro
References
3- usesA 20 ms onset rule left a mean 25.2 percent of shared STE and MNI marked time uncovered in the first hour of 18 SEEG sleep records sci:sha256:21d6910595e64dbef9553205b1617c944766bee1cc1d12004ffa856453c2ea99
We used its pinned ds005398 version 1.1.3 source archives, manifest, and retrieval code. We checked all source bytes against fresh public requests and made new agreement labels and prediction models.
- doi:10.1111/epi.18545
- usesdoi:10.18112/openneuro.ds005398.v1.1.3
The analysis used event, channel, participant, and EDF header data from OpenNeuro ds005398 version 1.1.3.
Cited by
3 publicationsPublic catalog entries on Sciscape.
- usesReleased HFO scores reduced detector agreement loss in nine SEEG patients excluded from logistic model fitting
We used its exact compressed source tables and EDF headers, pinned public manifest, unchanged parser and retrieval code, and prior results. We tested new logistic models on a fixed group of patients excluded from fitting.
- usesThe selected MNI count deficit within 25 ms of 10-second boundaries did not repeat in nine SEEG records
We used its pinned input archive for 18 SEEG sleep records from OpenNeuro ds005398 version 1.1.3 and copied its retrieval code. We checked all 94 source files or EDF headers against fresh pinned public bytes. We did not reuse its detector-agreement endpoint.
- usesAssuming independent, exchangeable patients, exact 1 s rotation correction gave an uncertain SOZ AUC gain for guarded 20 ms HFO matches in 18 SEEG records
We reused the pinned source archive, manifest, and retrieval code for the 18 SEEG records. This enabled offline analysis and fresh public-byte checks.